Monologue: Anthropic Slows Down

In progressEd Zitron (Better Offline)podcast2026-10-16read planted ai llmsbusiness

Synopsis — AI-drafted from Dan's notes

A 14-minute solo episode of Better Offline, aired on 16 October 2026. Ed Zitron covers the same week as his Tech Report interview, at monologue speed.

Anthropic’s third-quarter numbers leaked from investor meetings ahead of its IPO: $16.8B in revenue, roughly 45% up on the quarter. The online crowd had expected $20B or more, because the pitch had always been near-vertical growth. Anthropic also gave no new annualised run rate. Zitron explains it as a short period’s revenue multiplied up, often a trailing 28 days times 13, and calls it “deeply deceptive”. The last figure was $65B at the end of July, and talk of $100B by year end has now gone quiet. The adjusted margins (85% gross, above 11% operating) leave out costs such as stock compensation and are not GAAP. He also notes that Anthropic told investors 5–7% of its revenue came from Chinese labs distilling its models, roughly $0.8–1.2B of the quarter, and that it has since cut that sharply. By his arithmetic, keeping pace would need about $25.8B next quarter, and anything under $20B would be a disaster.

Around that he gathers the rest of the week. Ramp’s data shows 80% of spending on OpenAI and Anthropic comes from 1% of firms. CoreWeave’s debt yields about 15%, while Jensen Huang said Nvidia needn’t backstop it. Gartner expects 80% of neoclouds to stop operating independently by 2030. Jefferies warned of capital destruction. And the $413B in commitments behind Anthropic’s valuation depends on growth continuing.

The second half is a declaration of tone. Zitron says the show had lost its fun because he’d held back his contempt, and that contempt is now the show. Any journalist who prints run rates without a warning “like a pack of cigarettes” is failing at the job, he says, and one who thinks they’re a useful measure is “either ignorant or corrupt”. The more real numbers come out, he says, the worse the economics of AI look. He closes by trailing an interview with Tom Dotan.

Connections